I woke up in the middle of the night a few days ago and discovered that OpenAI had finally released those mathematical “results” they had been bragging about lately. We are talking about 722 (seven hundred and twenty-two) “papers” written by their private LLM or something. They apparently decided to do so after consulting the Advisory Group on Mathematics and Artificial Intelligence, a group of nine people who, frankly, could do better things for themselves and for society with their own time (like actual research, or teaching). I do not really care about the details, either of the secret OpenAI LLM, or the AGMAI or, perhaps surprisingly to some colleagues, of the “results” themselves.
I am writing this on the occasion of the OpenAI release, as I am now driven by a sense of urgency, but I have been planning to write a statement on AI in academia for a long time. I believe that we, as researchers and teachers, must reject AI-powered “papers” like these as legitimate research, because they are not. They do not follow any of the rules or the values of academia: building on the work of others with citations, not scooping your colleagues, collaborating, learning, teaching… and, you know, actually writing your own stuff in good faith. And even if they were legitimate research, it is research that is not worth doing: its negative side effects are just too many. We have a right not to read papers like these, we have a right and, I believe, a duty, not to cite them, and to refuse refereeing them.
Even if you, as a senior academician, believe that it is OK for yourself to do AI-powered “research”, you should first think about your students and your junior and future colleagues. AI-powered “research” is not self-sustaining: you will always need a human expert in the loop, even if it is just for writing prompts for your LLM chatbot, filtering all the bullshit answers, and interpreting and applying the results (a job that, frankly, I would not wish on my worst enemy). If AI-powered “research”, let alone “teaching”, becomes the norm, we will lose at least a generation of human experts. I have personally witnessed whole classes of students pressured (by peers and, above all, by the hypocritical behaviour of their teachers and the university bureaucrats) into using LLMs to write code they do not understand. These students are not learning. They will never be the human expert in the loop. And the same thing will happen to our junior colleagues, if you force them to use LLMs to do their job for fear of missing out. In fact, if we go on this way, one day there will be no expert at all. Even current experts will one day lose their expertise too, if they let LLMs do their science for them. Snap out of it.
And those colleagues who are actually doing research on LLMs as generators (even though I fear that I should probably say on LLMs tout court by now)… Yes, unfortunately you are part of the problem. Do something else. There is a world of research to be done that is actually useful, or at least not detrimental, to humanity. Yes, it is hard to change your research area, particularly when there is money in it, but I do not think it is ethically justifiable to contribute to the destruction of academia anymore.
Acknowledgements and AI statement. I would like to thank Pierre Clairambault, Barbara Fantechi, Maxime Folschette, Joshua Grochow, Lê Thành Dũng (Tito) Nguyễn, and Athénaïs Vaginay for their encouragement. Needless to say, no AI at all was involved in the preparation of this text (or anything else in my work).
Received 10 October 2026, revised 27 October 2026.
Add to the discussion